Most people's first impression of AI is shaped by a lazy prompt and a mediocre answer. The gap between "this is useless" and "this saved me an hour" is almost always the instructions you give it. Prompting isn't a dark art — it's clear communication, and you already know how to do that.
The anatomy of a good prompt
Think of it like briefing a sharp new assistant who can't read your mind. The best prompts include four things:
- Role & goal: "You're helping me, the owner of a 15-person landscaping company, write..." Tell it who it's being and what you're trying to achieve.
- Context: the facts it needs — your audience, your tone, relevant details, the actual source material. AI knows nothing about your business unless you tell it.
- The task, specifically: not "write something about our services" but "write a 150-word email to past customers offering a spring cleanup, friendly and direct, with one clear call to action."
- Format: "Give me three options," "use bullet points," "keep it under 100 words." Tell it the shape you want back.
A weak prompt: "Write a job posting." A strong prompt: "Write a job posting for a full-time HVAC service technician in Kansas City. We're family-owned, value reliability over flashiness, pay competitively, and offer a take-home truck. Warm but professional tone. Include responsibilities, requirements, and how to apply. Under 350 words." Same tool, completely different result.
Context is the cheat code
The single biggest lever is feeding the AI your own material. Paste in the email thread, the policy document, the rough notes, the customer's question. AI is dramatically better at working with information you provide than at recalling facts on its own. "Summarize this and draft a reply" beats "write me a reply about returns" every time — because now it's grounded in reality instead of guessing.
Iterate — don't expect one shot
The first answer is a draft, not a verdict. The pros treat it as a conversation: "Good, but make it warmer." "Shorten the second paragraph." "Add a line about our warranty." Each correction steers the result. You'll get to a great answer in three quick exchanges far more often than in one perfect prompt. If an output is way off, don't just retry — tell it what was wrong.
Where AI gets things wrong
Remember the hallucination problem from Module 1: an LLM can produce false information with complete confidence. It happens most when you ask for specific facts it wasn't given — a statistic, a citation, a price, a law, a person's details. It doesn't know it's wrong; it's just predicting plausible-sounding text. Watch especially for:
- Invented numbers, dates, quotes, or sources.
- Confident answers about your specific policies it was never told.
- Out-of-date information (models have a knowledge cutoff).
- Subtle errors buried in otherwise-correct text.
Verify before you ship — the rule of thumb
Match your scrutiny to the stakes. A brainstorming list for your eyes only needs little checking. Anything going to a customer, into a contract, onto your website, or near money or legal matters gets a human review — every time. The reliable pattern is AI drafts, a human decides. You keep the speed and stay accountable for what goes out the door. This is the same principle behind the human-in-the-loop approach we'll cover in the governance module. If you want your team fluent in this, structured AI training turns it into a habit. New terms tripping you up? Keep the glossary handy.
Want your team writing prompts that actually deliver? A free 20-minute AI Quick Wins call can pinpoint the few prompts that would save your people the most time this month.